Active Object Recognition Using Mutual Information

摘要: In this paper, we present the development of an active object recognition system. Our system uses a mutual information framework in order to choose optimal sensor configuration for recognizing unknown object. System builds conditional probability density functions database some observed features over discrete set configurations interesting objects. Using sequential decision making process, our determines action (sensor configuration) that augments discrimination between objects database. We iterate procedure until about class can be made. Actions include pan, tilt and zoom values camera. Features color patch mean region image. have tested on composed 8 different soda bottles obtained rate 95 %. Sequential length was 4 actions average